Comparative study of WaveNet-IIR PID controllers applied to a 2 GDL helicopter

In intelligent control applications, one problem is determining the number of layers and neurons in each layer. The problem becomes even more complex when the neural network includes functions like WaveNets, where translations and dilations are additional parameters. This article presents a comparat...

Descripción completa

Detalles Bibliográficos
Autores: Garcia-Castro, Oscar Federico, Ramos-Velasco, Luis Enrique, Garcia-Rodriguez, Rodolfo, Vega-Navarrete, Mario Alejandro, Escamilla-Hernández, Enrique, Oliva-Moreno, Luz Noe
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2022
País:México
Institución:UNIVERSIDAD AUTÓNOMA DEL ESTADO DE HIDALGO
Repositorio:PÄDI Boletín Científico de Ciencias Básicas e Ingeniería del ICBI
Idioma:español
OAI Identifier:oai:repository.uaeh.edu.mx:article/10067
Acceso en línea:https://repository.uaeh.edu.mx/revistas/index.php/icbi/article/view/10067
Access Level:acceso abierto
Palabra clave:Intelligent control
Identificacion for control
Linear adaptive control
Adaptive control by neural networks
Controladores inteligentes
Identificación para control
Control adaptativo lineal
Control adaptativo por redes neuronales
Descripción
Sumario:In intelligent control applications, one problem is determining the number of layers and neurons in each layer. The problem becomes even more complex when the neural network includes functions like WaveNets, where translations and dilations are additional parameters. This article presents a comparative study to determine the type of wavelet and number of neurons that best perform to approximate the Quanser helicopter dynamics with two degrees of freedom (DOF). The identification is based on a radial-based neural network whose activation functions are wavelets together, a pair of infinite impulse response (IIR) filters to ``prune'' some neurons. Additionally, a PID-WaveNet-IIR is presented, composed of a set of discrete PID controllers with self-tuning gains. Through numerical simulations using LabVIEW, the performance of the closed-loop system is presented under different operating conditions, types of family wavelets, where the minimum values of tracking errors are given previously, the number of neurons in the network, and the number of IIR filter lead and lag coefficients.